Sitepeep Β· Conversion Audit

Notion

Notion has built an excellent AI activation engine (five minutes from signup to a working relational database), then hidden it behind three screens of its own commercial priorities, and forgotten to show the user what it built.

πŸ”Ž Live traversal Β· notion.so/signup πŸ‘€ Onboarded as Founder, small team πŸ†“ Free tier, fresh workspace 🎨 Tokens pulled from live DOM βœ‰οΈ 1 activation email captured πŸ“… 21 Jul 2026
01: Executive Summary

The product delivers its promise in five minutes. The first-run sequence spends four of them selling.

A brand-new account, walked end to end on the free tier as a founder setting up a shared team workspace: the exact profile most likely to convert and most likely to leak.

5min
signup β†’ a real relational database, seeded and working. The strongest asset in the funnel.
4
first-run moments that read as failure: padlocked starts, an invisible reveal, an empty Home, unusable mobile web
15
prioritised recommendations, each citing a named principle
13
wireframes drawn in extracted Notion tokens, one per recommendation

Notion's AI workspace builder is genuinely very good. Answer one question about your role, press a button labelled "Build it", and roughly forty seconds later the sidebar contains Projects, Tasks, Meetings and Docs: real databases, with real relations between them, seeded with plausible starter rows. For a category whose historic failure mode is the paralysing blank page, this is a serious piece of product engineering. It resolves the oldest complaint about Notion almost entirely.

The problem is everything wrapped around it. Between the account existing and that engine starting, a new user is asked to invite their teammates, choose a paid plan, and download the desktop app: three consecutive screens, none of which serves the user, all of which serve Notion. The user has at this point seen nothing the product can do. They are being asked to recruit, to pay, and to install, on the strength of a brand promise alone.

The defining moment. After the AI finishes building, it writes: "All set. Let me show you what's been built." Then it shows nothing. The canvas stays empty. Beneath the sentence sit three text links (Open Meetings, Check out Tasks, Start with Docs) inviting the user to go and look for the payoff themselves. The single highest-emotion moment in the entire funnel, the one the whole sequence was built to produce, is delivered as a paragraph.

That miss compounds at the other end. Notion's Home, the surface a returning user lands on, the thing that decides whether day two happens, is empty. Not sparse: empty. The words "Welcome home" on white space, for an account that had just had four populated databases built into it minutes earlier. Above them sits a banner announcing a sidebar redesign to a user six minutes old who has no previous sidebar to compare it to.

The cost, stated plainly. Notion built a machine that manufactures the aha moment on demand, then declined to show the user the output, and gave them nothing to come back to. Two changes, rendering the artefact at the reveal and making Home reflect what exists, would convert an activation engine that already works into one the user can actually feel working.

The rest is smaller but consistent in direction. Two of the three suggested starting points on the first screen inside the product are padlocked, gated behind the plan the user declined one screen earlier. The free plan's feature list is set in a grey that fails accessibility contrast standards. The plan comparison describes the free tier as "1000 blocks" to a person who has not yet learned what a block is.

None of this is incompetence. It reads as a funnel that has been optimised one screen at a time, each addition locally justified, with nobody accounting for the cumulative cost to the person walking through all of them in sequence. That is the most common way good activation funnels rot, and it is very fixable.

02: Style Guide

A disciplined, deeply-tokenised system, with two measurable contrast failures on the money screen.

All values below were read from the live DOM during traversal: 728 CSS custom properties resolved from getComputedStyle, not sampled from screenshots or taken from documentation.

Text / Primary
#2C2C2B
Text / Secondary
#7D7A75
Text / Tertiary
#A19E99
Accent / Blue
#2383E2
Surface / Secondary
#F9F8F7
Border / Primary
#E6E5E3

Typography: system sans stack

Choose your Plan
Section heading Β· 650 semibold
Body copy runs at 16px / 24px line-height.
Mono label Β· 12px Β· tracked

Stack: ui-sans-serif, -apple-system, system-ui, "Segoe UI Variable Display", Helvetica, Arial. No webfont is loaded for UI chrome, a deliberate performance choice that renders natively per-platform. Weights 400 / 500 / 650 / 800. Buttons run 14px / 500.

Shape, elevation & components

Radius 8px buttons12px cardsPill chips Padding 0 12pxSidebar treeSlash menu Inline DB tablesDocked AI panelQuest progress card Template preview cardsEmoji page icons
⚠ Two measured contrast failures, both on the plan screen. Free-tier feature text renders #A19E99 on white at 14px, a ratio of 2.66:1, against a WCAG AA requirement of 4.5:1. The primary blue CTA renders #F3F9FD on #2783DE at 3.72:1, also below AA. The second is systemic: that button style recurs throughout the product.
On the token architecture itself. Notion's palette is unusually rigorous: ten colour families, each with a full tonal ramp and separate semantic slots for text, icon, border and background, plus dedicated alpha variants for layering. It is the work of a team that takes design systems seriously, which is precisely why the two AA failures stand out: this is a system fully capable of expressing an accessible grey, using an inaccessible one at the moment it asks for money.
03: Friction Log

Screen by screen, in the order a new user meets them.

Severity is assigned against one bar only: does this help a brand-new user reach a genuinely useful artefact quickly, and come back tomorrow? Where a choice appears to be a deliberate design decision rather than a defect, it is logged as a trade-off, with the caveat that intent is inferred from outside, not confirmed.

Major Friction Trade-off (documented decision) Keep: it's working

Signup /signup

Keep
A single field and no password. Email, or Google / Microsoft SSO. About as close to frictionless as an account gate gets, and correctly treated as pure cost to be minimised.
Keep
The work-email prompt explains itself: "Use your work email (if you have one) so it's easier for your team to join you on Notion." A request with a reason attached, tied to the team outcome. This is the single best piece of persuasive copy in the entire flow.
Friction
F-SU1That rationale is never repeated where it matters. The signup screen explains why teammates matter; the invite screen two steps later, the one actually asking for teammates, offers no reason at all. The argument exists in the product, at the wrong moment.
Trade-off
F-SU2The modal sits over a blurred grey skeleton of the app rather than real content. It signals "there is a product behind this" without promising anything specific, a reasonable choice, though a genuine populated workspace behind the glass would do far more to justify the cost of signing up.
Friction
F-SU3Apple SSO is absent: only Google and Microsoft. A defensible read of a work-first audience, but it removes an option many individual users default to.

Intent & personalisation /onboarding Β· "How will you use Notion?"

Keep
Three options: Work, Personal life, School. Clean against Hick's Law; no scroll, no sub-menu, decision cost near zero.
Keep
The answer visibly changes the product in real time. Selecting "Work" immediately swapped the workspace icon to πŸ’Ό and seeded a "Getting started" page in the sidebar behind the modal. This is the rare personalisation question that actually pays out rather than quietly harvesting a data point.
Friction
F-IN1That payoff is hidden behind the modal that caused it. The sidebar updates behind a centred overlay the user is looking away from. The single best piece of evidence that the question was worth answering is placed where it will mostly go unseen.
Friction
F-IN2Marketing consent is pre-checked and bundled into the same screen as progression. Opt-out-by-default consent on a screen whose primary action is "Continue" is a pattern that erodes trust cheaply, and sits awkwardly against opt-in consent regimes.
Trade-off
F-IN3No option is preselected, so Continue starts disabled. This protects the integrity of the segmentation signal at the cost of the Default Effect. A preselected "Work" would move more users forward faster. Defensible either way; the data is presumably worth more to Notion than the marginal step.

Invite teammates /onboarding Β· "Who else is on your team?"

Keep
The workspace and first teamspace are auto-named and auto-created, "Oliver Berger's Space" and "…Space HQ", with no naming chore. Excellent application of smart defaults to a step most competitors turn into a form.
Major
F-IV1The invite ask arrives at step two, before the user has seen anything. They are asked to spend social capital, to put their name behind this tool with colleagues, on the basis of a brand promise and nothing else. The Reciprocity Principle runs strictly one way here: the ask precedes any value given.
Friction
F-IV2There is no labelled skip. Leaving the fields empty and pressing "Continue" is the skip, but nothing says so. Two screens later the desktop-app step offers an explicit "Skip for now": the same action, given a clear affordance in one place and an ambiguous one here.
Friction
F-IV3No reason to invite is offered: no statement of what collaboration unlocks, no social proof, no preview of the shared surface. Three empty email fields and a copy-link button, presented without argument.

Choose your Plan /onboarding Β· plan selection

Major
F-PL1A pricing decision, at step three, before a single feature has been experienced. The user is asked to evaluate a $24/member/month plan against a free one with no basis for comparison beyond two bullet lists. This is monetisation placed inside the learning path.
Friction
F-PL2The free tier is described as "1000 blocks". "Block" is Notion's internal noun for its atomic content unit; a user three screens into their first session has no way to judge whether 1000 is generous or restrictive. The limit that most shapes the free experience is expressed in vocabulary the user has not been taught.
Friction
F-PL3The free option is visually suppressed. Business gets a solid blue CTA and full-colour feature icons; Free gets a pale tinted CTA and grey ticks. The Von Restorff Effect is deliberately pointed at the paid tier. That is legitimate commercially, but it makes the recommended path the one that costs money before value is proven.
Major
F-PL4Accessibility: free-tier feature text measures 2.66:1 against white, well under the 4.5:1 AA threshold. The design suppressing the free plan is also, measurably, the design making it hard to read.

Desktop app nudge /onboarding Β· "Notion is 50% faster with the Desktop App"

Keep
Explicit, clearly-labelled "Skip for now", and a specific, checkable benefit claim rather than vague enthusiasm.
Friction
F-DA1Still pre-value. This is a retention mechanic (install the app, form the habit) fired before the activation it depends on has happened. Under the Fogg Behavior Model, the trigger arrives when motivation is at its lowest point in the session: the user has now paid three screens of cost and banked nothing.

First landing /chat Β· "Welcome to Notion"

Trade-off
F-FL1The first surface inside the product is the Notion AI chat, full-bleed and with the sidebar collapsed, rather than a page or a Getting Started doc. This is a significant departure from the seeded-pages pattern Notion ran historically, and it is clearly intentional: AI is being positioned as the front door. The upside is a guided path instead of a blank canvas. The cost is that the workspace the user just watched being personalised is hidden the moment they arrive.
Major
F-FL2Two of the three suggested starting points are padlocked. "Never take notes again" and "Wake up to a clear inbox" render greyed with a lock icon, gated behind the plan the user declined one screen earlier. At the highest-stakes moment in the funnel, two-thirds of the offered paths are dead ends that exist to punish a choice the product just invited them to make.
Keep
The greeting is warm, time-aware and personal ("Good morning, Oliver. Welcome to Notion."), and the copy disarms AI fatigue rather than ignoring it: "I'm Notion AI. Yes, another AI, I know." Confident, human, and it earns attention rather than demanding it.

The AI setup wizard /chat Β· role question β†’ build

Keep
Closed-form questions, not open prompts. "What's your role?" answered by eight tappable chips rather than a text box. For a first-timer who does not yet know what the system can do, picking beats describing every time.
Keep
The answer is reflected back in the user's own language: "Lots to manage for a founder β€” deals, product decisions, team tasks, and the docs that hold it all together." Textbook Commitment & Consistency: the user's choice is restated as understanding before anything is asked of them.
Keep
The CTA reads "Build it β†’", not "Continue". A committal verb at the moment of commitment.
Friction
F-AW1Roughly 70% of the canvas sits empty throughout the wizard. Questions render in a narrow centred column with a vast white field beside them. That is exactly the space where a live preview of the workspace assembling itself would convert an interview into a co-creation.
Friction
F-AW2Generation is covered by a bare one-word spinner ("Brewing", then "Vibing") on an otherwise blank screen. The labels have personality, but there is no skeleton, no item-by-item progress, no sense of what is being made. Perceived Performance is left entirely on the table during the longest wait in the flow.

The reveal /chat Β· "All set."

Major
F-RV1"All set. Let me show you what's been built." And nothing is shown. The canvas remains empty; below the sentence are three text links to go and find the work elsewhere. The Peak-End Rule says users remember the emotional peak and the ending. This is the designated peak of the entire first session, and it renders as a paragraph of text. The artefact exists, is good, and is invisible.

First database /p/… Β· Meetings

Keep
This is the genuine aha, and it is strong. A real database: two seeded rows, typed columns, two saved views, and live relations linking each meeting to a Project. It demonstrates the one thing a document cannot do, without requiring the user to build it. Reached roughly five minutes after signup.
Keep
Notion AI redocks as a right-hand side panel beside the artefact and narrates it in context. The assistant and the work are on screen together, the co-presence the wizard itself was missing.
Friction
F-DB1The recommended first stop is Meetings, arguably the least immediately useful of the four for a solo founder on day one with no meetings recorded, and the one whose headline capability (AI meeting notes) is locked on the free plan. Tasks or Projects would demonstrate value with the seeded data actually present.

The blank canvas /p/… Β· new page & slash menu

Keep
"Add new" no longer opens a blank page. It opens a picker of visually-previewed templates (Tasks Tracker, Projects, Document Hub, Brainstorm Session) with "Empty page" and "Empty database" as explicit escapes. This is a direct, effective answer to the paralysis problem and a clear improvement on the historic pattern.
Keep
The slash affordance does announce itself, at the cursor, in the body: "Press 'space' for AI or '/' for commands." Both the AI and the block menu are advertised at the exact point of action.
Trade-off
F-BC1The hint is conditional. It appears only once the cursor enters the body, so on first arrival, with focus in the title, the page carries no guidance at all. Consistent with Notion's minimalism (the interface stays quiet until you engage), at the cost of a first-timer's very first second on the surface.
Friction
F-BC2The persistent "Get started with" bar sits 550px below the cursor, pinned to the viewport bottom. That figure is measured. The guidance is placed at maximum distance from the point of attention, which is precisely backwards under Fitts's Law. Its chips also name features ("Database", "Form") rather than outcomes.
Friction
F-BC3The slash menu's two top "Suggested" slots go to AI Meeting Notes (Beta) and HTML Embeds (New): Notion's newest features, not a first-timer's likely need. The highest-attention position in the most-used menu in the product is allocated to promotion rather than to Text, Heading or To-do list.
Keep
Below those two slots the menu is well-built: grouped by category, type-to-search from the first keystroke, keyboard shortcuts shown inline. Hick's Law is properly mitigated across 25+ options.

Home & the return experience /home

Major
F-HM1Home is empty. The words "Welcome home" on white space, with no recent pages, no tasks, no upcoming meetings, no progress and no next action, for an account that had four populated databases built into it minutes earlier. The surface that decides whether day two happens shows the user nothing they own and gives them no reason to return.
Friction
F-HM2Returning to the root URL restores the last page visited, not Home. Correct for an established user; for a new one it means day two opens on whatever half-finished scrap they abandoned. In this traversal, that was a near-empty page with a stray "/" still sitting in it.
Friction
F-HM3Home carries a banner announcing that "Home is getting an update in our new sidebar redesign", a migration notice served to an account six minutes old, which has no previous sidebar and no basis to care. Internal product-change comms leaking into the first-run experience.
Keep
The sidebar quest card persists ("Get organized in minutes", with a partial progress bar), carrying an incomplete task forward across sessions. Good use of the Zeigarnik Effect, and the one element on this surface doing retention work.

Template gallery app.notion.com/templates β†’ www.notion.com/templates

Major
F-TG1The in-app Templates link ejects the user out of the product onto the marketing site. They land on a page with full marketing navigation (Product, Solutions, Pricing, Enterprise, Request a demo) and a "Log out" button in the top right. A new user looking for a starting structure is now being sold to as a prospect, one click from ending their session.
Friction
F-TG2The page leads with paid services, not templates. The first and largest panel is "Work smarter with Notion experts", certified consulting partners. Below that, an AI Skills promo and a featured creator. A user who clicked "Templates" must scroll past three non-template units to reach templates.
Friction
F-TG3The page contradicts itself on its own core number: the document title says "30,000+ templates" while the search field on the same screen says "Search 70,000+ templates". Small, but it is the kind of detail that quietly costs credibility.
Friction
F-TG4Nothing is personalised to the answers already given. This user told Notion they are a founder setting up a team workspace, and the AI then built them Projects, Tasks, Meetings and Docs. The gallery greets them with an undifferentiated marketplace and a search box, a Hick's Law problem at the scale of tens of thousands of options.

Mobile web /home at 390 Γ— 844

Major
F-MW1At a standard phone viewport the app does not adapt. The sidebar holds a fixed 270px, 69% of a 390px screen, leaving a 120px content column. Home is effectively unreachable. Measured, not estimated.
Major
F-MW2Inside that 120px column, the sidebar-redesign banner reflows to roughly one word per line and consumes the entire 844px viewport height. The first thing a mobile-web visitor sees is a wall of single-word lines about a feature change they have no context for.
Trade-off
F-MW3Notion clearly steers mobile users to native apps, and the desktop-app prompt in onboarding is part of that strategy. Treating mobile web as a low-investment surface is a legitimate allocation choice, but it is not the same as shipping a layout that breaks, and the activation email's own CTA lands here for anyone who opens it on a phone.

Activation email "A helpful teammate, built into Notion" Β· ~15 min after signup

Major
F-EM1The email is entirely about Agents: "You assign the tasks. Agent does the work for you", CTA "Start building with Agent". Agents are a Business-plan feature. The primary call to action in the first activation email points a free user at something they cannot use. This is the padlocked-cards problem again, in a second channel.
Major
F-EM2It makes no reference to what the user actually did. This account had just had Projects, Tasks, Meetings and Docs built into it by Notion's own AI fifteen minutes earlier. The email mentions none of it. There is no "your workspace is waiting", no link back to the artefact, no acknowledgement that a workspace exists. The most powerful available return hook is simply unused.
Friction
F-EM3Six competing calls to action: one primary button, four video guides, and a "Check it out" guide link. Asked what single action this email drives, the honest answer is that it does not know. Hick's Law applies to email as much as to interfaces.
Friction
F-EM4It landed in Gmail's Promotions tab rather than the inbox, a deliverability outcome consistent with its heavily image-led, multi-CTA marketing construction. A behaviourally-triggered, text-led message tied to the user's own workspace would both read better and route better.
04: Recommendations

Fifteen changes, ordered by what they do to time-to-first-artefact and next-day return.

Each carries a named principle from the evaluative framework. Three matter more than the rest, R1, R2 and R13, because each governs a different moment the user decides whether Notion is worth continuing with: seeing the value, returning to it, and being pulled back to it.

R1

Render the artefact at the moment of the reveal

AI chat Β· fixes F-RV1
Current state
The AI writes "All set. Let me show you what's been built" and then displays nothing. The user is offered three text links and left to go and find the output on their own.
Why it matters
This is the emotional peak the entire preceding sequence exists to produce, and it is the one moment rendered as plain text. A user who does not click a link never sees what they were promised, and leaves the session believing the product did less than it did. The work is already done; only the display is missing.
Recommendation
Render live previews of the created databases inline in the chat, immediately beneath the "All set" message: Projects, Tasks and Meetings as real, scrollable mini-tables showing their seeded rows. Show the thing, then offer the links. The only acceptable version of "let me show you" is one that shows.
Principle applied
Peak-End Rule + Zero State Design Β· PM Wisdom: Magic Moment, Not Aha Moment. Define the wow moment as a screenshot. If you can't, the product doesn't have one.
Proposed state: the AI reveal message with three rendered database previews (Projects, Tasks and Meetings) shown inline as real tables beneath it, above a single primary button reading Open my workspace.
01 Β· reveal renders the artefact
R2

Rebuild Home as the surface that earns day two

Home Β· fixes F-HM1, F-HM2, F-HM3
Current state
Home shows "Welcome home" and nothing else, for an account with four populated databases. Returning to the root URL bypasses it entirely in favour of the last page visited. A sidebar-redesign migration banner sits above it.
Why it matters
Activation is not complete until the user comes back. Home is the only surface whose entire job is to make that happen, and it currently performs none of it. A new user's day-two experience is being decided by a screen that shows them nothing they own.
Recommendation
Populate Home with the user's own material: a "jump back in" row of recent pages, open tasks from the Tasks database, upcoming items from Meetings, and the setup-progress strip currently stranded in the sidebar. Suppress product-change banners for accounts under ~7 days old. For new users specifically, route the root URL to Home rather than last-page for the first several sessions.
Principle applied
Goal Gradient Effect + Zeigarnik Effect Β· PM Wisdom: Dashboard Moment. Lead with the user's own progress, not a feature tour.
Proposed state: a populated Home headed Welcome back, with three cards (Jump back in, Your tasks, Upcoming) above a workspace setup progress bar reading 2 of 4 done and an Invite a teammate button.
02 Β· Home reflects what exists
R3

Move the invite ask to after the first artefact exists

Onboarding sequence Β· fixes F-IV1, F-IV3, F-SU1
Current state
Teammates are requested at step two, before the user has seen anything the product does, with no stated reason and no preview of what would be shared.
Why it matters
An invitation is the user spending their own credibility on Notion's behalf. Asked before value, it is a request for a favour; asked immediately after the AI builds a working team workspace, it is an obvious next step with something concrete attached to it. For a team product, this is the single most valuable step in the funnel. It is currently placed where it converts worst.
Recommendation
Relocate the invite step to directly after the reveal, framed against the artefact: "Your team workspace is ready. Bring in the people who'll use it," with a thumbnail of the actual Projects board. Keep the copy-link option, and label the skip explicitly.
Principle applied
Reciprocity Principle Β· PM Wisdom: Workspace-Level Activation. Workspaces of one churn, but the invite has to be earned before it's asked.
Proposed state: a four-step onboarding sequence with role complete, then an emphasised card reading Your workspace is ready containing a table preview, followed by Invite your team and finally Choose a plan de-emphasised.
03 Β· value before the ask
R4

Take plan selection out of the learning path

Onboarding sequence Β· fixes F-PL1, F-PL3
Current state
A full pricing comparison at step three, before any feature has been used, with the paid tier visually favoured and the free tier suppressed.
Why it matters
Onboarding's job is user value; the upsell's job is company value, and merging them costs both. A user asked to price a product they have not used cannot make a good decision, so they default to free while absorbing the impression that the good version is the one behind the paywall, before seeing that the free version is genuinely capable.
Recommendation
Default new signups to Free silently and move the plan decision to the first moment a real limit is met: approaching the block cap, inviting past the seat threshold, reaching for a locked feature. Sell against experienced constraint, not anticipated benefit.
Principle applied
Reciprocity Principle + Default Effect Β· PM Wisdom: Don't Let Monetisation Creep Into Onboarding. Save upsell for the moment users hit a real limit.
Proposed state: an in-product upgrade prompt appearing at a real limit, with an almost-full progress bar, a short heading reading You are close to your limit, and See plans alongside Not now.
04 Β· plan choice at a real limit
R5

Never show a locked option as a primary starting point

First landing Β· fixes F-FL2
Current state
Two of the three suggested starts on the first screen inside the product are padlocked, gated behind the plan the user declined on the previous screen.
Why it matters
Under the Fogg Behavior Model, a trigger without ability produces no behaviour. It produces frustration. Two-thirds of the options at the highest-intent moment in the session are worse than unavailable: they function as a reprimand for a choice the product had just invited. The user's first act inside Notion is being told no.
Recommendation
Show free-tier users three starting points they can actually complete. Surface locked capabilities later, in context, at the point where the user is doing the adjacent thing, rather than as their opening menu.
Principle applied
Fogg Behavior Model (ability) + Von Restorff Effect Β· PM Wisdom: Speed-to-Value Over Feature Completeness.
Proposed state: three equally weighted, fully enabled starting cards reading Get organised, Draft a doc and Track your work, each with its own Start button and no padlocks.
05 Β· three starts a free user can finish
R6

Build the workspace in view, beside the questions

AI wizard Β· fixes F-AW1, F-AW2
Current state
Wizard questions render in a narrow centred column with roughly 70% of the canvas empty. Generation is covered by a one-word spinner ("Brewing", then "Vibing") with no indication of what is being made.
Why it matters
A user watching their workspace assemble itself as they answer is a participant; a user watching a spinner is a bystander. The empty right-hand canvas is the most valuable unused space in the funnel, and the wait it covers is the longest in the session. The two problems solve each other.
Recommendation
Split the wizard: questions left, live preview right. As each answer lands, render the corresponding databases materialising in place. Replace the spinner with per-item progress ("Creating Projects… Adding Tasks… Linking Meetings to Projects…") so the wait itself demonstrates the product's reach.
Principle applied
Endowed Progress Effect + Perceived Performance Β· PM Wisdom: Guided Wizards Beat Tooltip Trails. The side-by-side wizard that co-creates in real time drove Airtable's largest single activation lift.
Proposed state: a split wizard with the role question and its chips on the left, and a live workspace preview on the right showing the sidebar tree and a table whose lower rows are still rendering.
06 Β· the wizard co-creates in view
R7

Fix the two AA contrast failures

Plan screen & global buttons Β· fixes F-PL4
Current state
Free-tier feature text renders #A19E99 on white at 2.66:1. The primary blue CTA renders #F3F9FD on #2783DE at 3.72:1. Both fall below the 4.5:1 AA threshold for normal text.
Why it matters
The first is a legibility problem on the screen where the user makes a financial decision. The second is systemic: that button style recurs across the product, so the failure repeats everywhere the primary action appears. Notion's token system already contains compliant greys; this is a selection error, not a design constraint.
Recommendation
Move free-tier feature text to #7D7A75 (secondary, comfortably AA at 14px). Darken the primary button to approximately #1A6DC0, or set its label to pure white and increase weight. Add an automated contrast gate to the token pipeline so semantic pairings are checked at build time.
Principle applied
Aesthetic-Usability Effect Β· WCAG 2.2 AA (1.4.3) Β· PM Wisdom: Taste Leaks Through Every Interaction.
Proposed state: a side-by-side contrast comparison of the free plan feature list, the failing version at 2.66 to 1 marked with a red cross and the corrected version at 4.8 to 1 marked with a green tick.
07 Β· free-tier text raised to AA
R8

Describe limits in outcomes, not internal vocabulary

Plan screen Β· fixes F-PL2
Current state
The free tier is defined as "1000 blocks" to a user who has not yet been taught what a block is. "Basic forms" and "Basic sites" carry the same problem.
Why it matters
Feature nouns teach the product's taxonomy; outcome language teaches value. At the point of a pricing decision, a limit the user cannot interpret reads as risk, and unquantified risk pushes people toward inaction or away from the free tier for the wrong reasons.
Recommendation
Translate to something scaled to real use: "About 1,000 paragraphs, tasks and rows, enough for a small team's first few months", and defer the block concept to the moment the user approaches the limit, where it can be shown rather than defined.
Principle applied
Cognitive Ease Β· PM Wisdom: Don't Name Features in Onboarding. Show What They Solve.
Proposed state: a free plan card describing limits in plain outcomes such as Room for a small team and Upgrade when you outgrow it, rather than block counts.
08 Β· limits described as outcomes
R9

Make every skip look like a skip

Onboarding sequence Β· fixes F-IV2
Current state
The invite step's skip is an unlabelled "Continue" with empty fields. Two screens later the desktop-app step offers an explicit "Skip for now" for the equivalent action.
Why it matters
Inconsistent affordances for the same action force the user to work out the interface rather than use it, and an ambiguous exit reads as a step that cannot be avoided. Users who feel trapped in a flow abandon it rather than complete it.
Recommendation
Use one skip pattern across every optional onboarding step: a consistently placed, consistently worded "Skip for now", and reserve the primary CTA for the action the screen is actually about.
Principle applied
Locus of Control Support + Cognitive Ease.
R10

Point the slash menu's top slots at first-timers

Slash menu Β· fixes F-BC3
Current state
The two "Suggested" slots, the highest-attention position in the most-used menu in the product, are occupied by AI Meeting Notes (Beta) and HTML Embeds (New).
Why it matters
For a user in their first session, those two entries are the least likely to be wanted and, in the case of AI Meeting Notes on free, not usable at all. The slot is doing feature marketing at the exact moment the user needs help composing their first block.
Recommendation
Make "Suggested" account-age- and plan-aware. In early sessions, populate it with the blocks first-timers actually reach for (To-do list, Heading, Table) and promote newer capabilities once the user has demonstrated baseline fluency.
Principle applied
Anticipatory Design + Default Effect Β· PM Wisdom: Opinionated Defaults Over Choice-Paralysis.
Proposed state: a slash menu whose Suggested section contains To-do list, Heading and Table, with newer features demoted below the divider.
09 Β· suggested slots serve first-timers
R11

Bring the starter actions to the cursor

Empty page Β· fixes F-BC2
Current state
The "Get started with" bar is pinned to the bottom of the viewport, a measured 550px below the cursor, and its chips name features rather than outcomes.
Why it matters
Guidance placed at maximum distance from the point of attention is guidance most users never register. The inline hint at the cursor already works well; the bottom bar duplicates that job from the worst possible position, and disappears on subsequent visits anyway.
Recommendation
Fold the starter actions into the cursor line as inline ghost chips on the first empty block of a new user's first few pages, phrased as outcomes: "Track something", "Write a doc", "Ask AI to draft this". Retire the bottom bar.
Principle applied
Fitts's Law + Zero State Design.
Proposed state: an empty page with three ghost chips reading Track something, Write a doc and Ask AI sitting inline at the text cursor, directly under the title.
10 Β· starter actions at the cursor
R12

Time the app and habit nudges to earned value

Onboarding sequence Β· fixes F-DA1
Current state
The desktop-app prompt fires at step four, before any value has been delivered. The setup-progress quest card, the flow's best retention device, is confined to the sidebar.
Why it matters
Install prompts convert on the back of demonstrated value. Fired pre-value they consume goodwill at the session's lowest motivation point; fired after a user has a workspace they like, the same prompt is a natural next step.
Recommendation
Move the app nudge to the end of the first successful session, or to the start of the second. Promote the quest card to Home (see R2) with its remaining steps named, so the incomplete-task pull is visible rather than tucked into the sidebar.
Principle applied
Fogg Behavior Model (trigger timing) + Goal Gradient Effect.
R13

Rewrite the activation email around what the user actually built

Lifecycle email Β· fixes F-EM1, F-EM2, F-EM3, F-EM4
Current state
The first activation email, arriving ~15 minutes after signup, is entirely about Agents, a Business-plan feature this free user cannot access. It makes no reference to the workspace Notion's own AI had just built, carries six competing CTAs, and landed in Gmail's Promotions tab. Whatever follows it was outside this audit's observation window.
Why it matters
This is the opening move in pulling a lapsed new user back, the touch with the most attention and the most context available to it, and it is pointed away from everything that would work. The user has a populated workspace with their name on it; the email talks about a feature they would have to pay for. It repeats, in a second channel, the exact pattern that undermines the first-run: selling the next tier before confirming the value of the current one.
Recommendation
Trigger on behaviour, not on a timer, and lead with the artefact: subject line naming their own workspace, body showing the databases that exist, one CTA returning them to it. Branch on what actually happened: built-and-left, built-and-returned and never-started are three different messages. Above all, keep tier-locked features out of the primary CTA for free users: whatever else the lifecycle programme does, its first touch should not open on something the recipient cannot use.
Principle applied
Zeigarnik Effect + Endowment Effect ("your workspace") + Hick's Law on CTA count Β· PM Wisdom: Don't Name Features. Show What They Solve.
Proposed state: an activation email headed Your workspace is ready, showing a preview of the built database above a single button reading Open my workspace, with no competing links.
11 Β· email leads with the artefact
R14

Keep the template gallery inside the product

Templates Β· fixes F-TG1, F-TG2, F-TG4
Current state
The in-app Templates link navigates to the marketing site, complete with Pricing, Request-a-demo and a "Log out" button, and leads with paid consulting rather than templates. Nothing reflects the role or workspace the user already declared.
Why it matters
A user hunting for a starting structure is in a high-intent build moment. Ejecting them into an acquisition surface breaks that intent, re-pitches them as a prospect they no longer are, and puts a logout control on screen. It also discards the segmentation signal the product spent two screens collecting.
Recommendation
Render the gallery in-app with product chrome intact. Open on a short, role-matched shelf ("Templates for founders") drawn from the signup answers, with the full marketplace one click deeper. The in-app picker already surfaced by "Add new" is the right pattern; extend it rather than routing users away from it.
Principle applied
Hick's Law + Anticipatory Design Β· PM Wisdom: Speed-to-Value. Deep-link users to the action, not to a catalogue.
Proposed state: an in-app template picker headed Templates for founders showing Fundraising, Hiring and Roadmap, with a quiet Browse all templates link and no marketing navigation.
12 Β· gallery stays in-product
R15

Make mobile web survivable, or route away from it honestly

Mobile web Β· fixes F-MW1, F-MW2
Current state
At 390px the sidebar holds 270px, 69% of the screen, leaving a 120px content column, and the product-change banner fills the full viewport height at roughly one word per line.
Why it matters
Deprioritising mobile web is a defensible strategic choice; shipping a broken layout is not the same thing. Every email CTA, every shared page link and every search result opened on a phone lands here, including the activation email's own primary button. A user whose first impression of Notion is this screen has been lost by a surface nobody intended them to see.
Recommendation
At minimum, collapse the sidebar below a breakpoint and suppress product-change banners on narrow viewports. Both are small, contained fixes. If mobile web is genuinely not a supported surface, say so with an honest interstitial offering the app or a read-only view, rather than rendering a layout that simply fails.
Principle applied
Aesthetic-Usability Effect + Locus of Control Support Β· PM Wisdom: Taste Leaks Through Every Interaction.
Proposed state: a phone screen with the sidebar collapsed behind a menu icon, a Welcome back heading and two readable content cards filling the full width.
13 Β· mobile web that fits the screen
04b: Secondary Lens Cross-check

The same flow, read against how strong PM teams judge activation.

Sharpening pass. These are observations rather than separate recommendations. Where they change a priority, they are folded into the numbered set above.

Major Friction Keep: it's working

Against the activation checklist

Keep
"How many steps from signup to first value?" The AI path answers this better than almost any product at Notion's complexity. One question, one button, ~40 seconds of generation, a working relational database. The engine is not the problem.
Major
"What is the wow moment, and can you sketch it as a screenshot?" This is the test Notion fails. The screenshot of the wow moment is a paragraph of text on an empty canvas. Everything needed to make it a real image already exists in the account; it simply is not rendered. Drives R1.
Major
"Is the home dashboard about the user's north star, or the product's feature tour?" Neither. It is blank. For a multi-user product where workspace-level activation is the real metric, the surface that should show collaborative momentum shows nothing at all. Drives R2.
Friction
"Does onboarding name features, or solve problems?" Mixed. The AI wizard speaks in outcomes ("track everything you're juggling"); the plan screen and the empty-page bar speak in nouns ("1000 blocks", "Database", "Form"). The good instinct exists in the newest surface and has not propagated. Drives R8, R11.
Friction
"Does workspace-level activation match user-level activation?" The traversal ends with an activated user inside a workspace of one. The invite step ran before there was anything to share and was trivially skippable, so the most likely outcome is precisely the pattern that predicts churn: a green individual metric over an empty workspace. Drives R3.
Keep
"Guided wizard or tooltip trail?" Firmly a wizard, and a well-built one: closed-form choices, the answer reflected back, a committal CTA. It is missing only the live co-creation panel that made the pattern famous. Drives R6.
Friction
"Is monetisation kept out of the learning path?" No. Pricing at step three, and its consequences reach forward into the product: the padlocked cards on the first screen inside Notion are the plan screen's aftershock. This single sequencing decision generates two separate Major findings. Drives R4, R5.
Major
"What single action does each lifecycle email drive, and is it tied to what the user did?" Neither. The first activation email carries six CTAs and its primary button points a free user at a Business-tier feature, while ignoring the workspace Notion had just built for them fifteen minutes earlier. Drives R13.
Friction
"Where does the product hand the user off to marketing?" At Templates, mid-build, onto a page selling consulting and demos with a logout button in the corner. High-intent users are re-cast as prospects. Drives R14.
Keep
"Are error and system messages human?" Yes, notably so. "Brewing", "Vibing", "Yes, another AI, I know". The voice is confident and self-aware without being cloying. The writing is a genuine asset; it is the rendering around it that underdelivers.
The pattern underneath all of it. Notion's newest surface, the AI wizard, embodies almost every principle a strong PM team would ask for. The surfaces around it, built earlier and optimised separately, embody the opposite. This is not a product that lacks the instinct; it is a funnel where the good instinct has not yet been allowed to set the sequence.
05: Implementation Roadmap

Prioritised by activation impact per unit of effort.

Effort: S hours–1 day Β· M days Β· L 1–2+ weeks. Impact β˜…β€“β˜…β˜…β˜….

Quick wins ship this sprint

#FixEffortImpactWhy now
R5Remove padlocked cards from free-tier first-run suggestionsSβ˜…β˜…β˜…Two of three first actions currently fail; config change, no new surface
R7Raise free-tier text and primary-button contrast to AASβ˜…β˜…Compliant tokens already exist; systemic fix, one-line change
R8Rewrite "1000 blocks" and tier bullets as outcomesSβ˜…β˜…Copy-only; sits on the screen where money is decided
R9Unify the skip affordance across onboarding stepsSβ˜…Copy and layout only; removes a false forced-step signal
R12Move the desktop-app prompt to end of first sessionSβ˜…β˜…Pure sequencing; recovers a pre-value interruption
R2Suppress product-change banners for accounts < 7 days old (partial R2)Sβ˜…Audience rule on an existing banner system; separable from the Home rebuild
R15Collapse the sidebar below a mobile breakpoint; hide banners on narrow viewportsSβ˜…β˜…Two contained CSS fixes turn an unusable screen into a survivable one
R13Remove tier-locked features from the free-user email's primary CTASβ˜…β˜…Copy and audience targeting only; the current CTA points free users at a paid feature

Medium next 2–4 weeks

#FixEffortImpactWhy
R1Render created databases inline at the "All set" revealMβ˜…β˜…β˜…The single highest-value change in the audit; inline DB previews already exist elsewhere in the product
R3Relocate the invite step to after the reveal, framed on the artefactMβ˜…β˜…β˜…Highest-value step for a team product, currently placed where it converts worst
R10Make slash-menu "Suggested" account-age and plan awareMβ˜…β˜…Highest-attention slot in the most-used menu, currently spent on promotion
R11Replace the bottom starter bar with inline chips at the cursorMβ˜…β˜…Moves guidance 550px closer to where attention already is
R13Rebuild the first activation email around the user's own workspace, behaviourally triggeredMβ˜…β˜…β˜…The opening return touch currently points away from everything that would work
R14Bring the template gallery in-app with a role-matched opening shelfMβ˜…β˜…Stops ejecting high-intent users onto a marketing page with a logout button

Strategic quarter horizon

#FixEffortImpactWhy
R2Rebuild Home around the user's own content and progressLβ˜…β˜…β˜…The surface that decides day two currently does no retention work at all
R6Side-by-side wizard with live workspace preview and per-item progressLβ˜…β˜…β˜…Turns the interview into co-creation; the pattern's largest documented activation lift
R4Default new signups to Free; move plan choice to first real limitM–Lβ˜…β˜…Removes monetisation from the learning path; needs billing and lifecycle coordination
If only three things ship. R1, R2 and R13: see it, return to it, be pulled back to it. One makes the user see the value that already exists; one gives them somewhere to come back to; one stops the opening activation email pointing them at a feature they cannot use. Every other item here improves a funnel. Those three decide whether the user remembers that Notion built them something.
06: Success Metrics

What to instrument, and what would count as the changes having worked.

Targets are framed as directional deltas rather than absolutes. An outside auditor has no access to Notion's baselines, and a specific number quoted without one would be fabrication.

Time-to-first-useful-artefact

  • Median minutes from signup to a database or page containing real user content, excluding the AI's seeded rows. The traversal reached a seeded database in ~5 minutes; the meaningful question is when the first user-authored row appears.
  • Reveal-to-artefact-open rate: share of users who click through from "All set" to an actual database. Under R1 this becomes a measure of whether the preview is compelling rather than whether the links are noticed.
  • Track by first-session cohort rather than aggregate. Velocity to first milestone is the leading retention indicator.

Activation rate

  • Define against a retention-correlated threshold, not the easiest reachable milestone: e.g. 3+ user-created artefacts across 2+ sessions in the first 7 days.
  • Validate the chosen threshold correlates at least 2Γ— with 30-day retention before optimising against it. Resist lowering the bar to make the number move.
  • Report separately for the AI-wizard path and the manual/template path; they are different products in practice.

Template-to-active conversion

  • Share of AI-built or template-created databases that receive a user edit within 7 days. This is the honest test of whether generated structure becomes used structure, or decoration.
  • Per-database breakdown: Projects vs Tasks vs Meetings vs Docs. F-DB1 predicts Meetings underperforms for solo founders; the data settles it.
  • Deletion rate of seeded content as the counter-metric: scaffolding that gets binned is clutter, not a head start.

D1 / D7 return

  • D1 and D7 return by signup cohort, split by whether the user reached an artefact in session one. The gap between those two curves is the size of the activation prize.
  • Home engagement on return: proportion of day-two sessions where the user acts on something surfaced by Home. Currently near-unmeasurable, since Home surfaces nothing. This metric only becomes real after R2.
  • Quest-card completion as a leading indicator of the incomplete-task pull actually working.

Invite & collaboration rate

  • Invite-send rate at the current step-two position vs. the post-reveal position in R3, the cleanest available A/B in this whole audit.
  • Invite acceptance rate, not just sends: an invitation attached to a real workspace should convert far better than one attached to a promise.
  • Workspaces with 2+ weekly-active contributors at day 30, the workspace-level metric that individual activation hides. Workspaces of one churn regardless of how green the user metric looks.

Guardrails

  • Free-to-paid conversion must be watched through R4. Moving plan choice out of onboarding should be validated as neutral-to-positive on 90-day conversion, not just on activation.
  • Block-limit encounters: how many free users approach 1000 blocks, and what they do next. The upsell moment R4 relies on has to actually occur.
  • AI acceptance rate, the share of generated artefacts kept and edited rather than deleted. For a product now leading with AI, this is the number that decides whether the front door is load-bearing.
North-star framing. For a collaborative workspace, the honest north star is workspaces with multiple contributors and recurring weekly use, not activated users. Every recommendation here is ultimately in service of that: show the user what was built (R1), give them a reason to return (R2), and ask them to bring someone else only once there is something worth sharing (R3).

See your funnel this clearly

Want this depth on your own funnel?

This audit named 15 conversion leaks in Notion's activation journey, each one diagnosed, prioritised, and anchored to a behavioural principle. Sitepeep does the same for your product: every leak across your funnel, a prioritised roadmap, and the science behind each fix.